AI Frontier · Aug 14, 2026
This section is a 7-day rolling roundup, not a daily snapshot — AI vendor blogs typically publish weekly, and a 36-hour window is frequently empty, which is precisely why this section used to come up empty. Known gap: the three
ai_securitysources (HiddenLayer Research / Lakera Blog / Protect AI Blog) currently have no usable RSS and are permanently 0 rows inraw_items; that does not mean they published nothing. See the repository CLAUDE.md for details. Anthropic News / Anthropic Research also had no new entries ingested within this window.
Last 48 Hours
- [Aug 13] Google DeepMind releases Gemini 3.7 Flash: the next generation of its lightweight model. (Link)
- [Aug 13] OpenAI publishes "The builder's guide to GPT-5.6": aimed at startups, covering how to build faster, more cost-efficient AI agents through smarter model selection and new Responses API capabilities. (Link)
- [Aug 13] OpenAI previews the Ultrafast service tier: GPT-5.6 Sol, running on Cerebras compute, reaches up to 14x the speed with output rates as high as 750 tokens per second. (Link)
- [Aug 13] Google launches Sheets Canvas: turns Google Sheets data into interactive dashboards, study trackers, and more via prompts. (Link)
- [Aug 12] Google DeepMind releases SL2T, a sign language recognition model: a breakthrough sign-language-to-text model for deaf and hard-of-hearing users, already shipping in new sign language features. (Link)
Earlier Than 48 Hours (Within the 7-Day Window)
- [Aug 12] OpenAI publishes research on enterprise agentic AI adoption: how enterprises are adopting ChatGPT and Codex, with frontier firms pulling further ahead of the rest of the field on AI adoption. (Link)
- [Aug 11] OpenAI expands Daybreak cybersecurity capabilities: releases GPT-5.6-Cyber to approved Daybreak Red partners for authorized vulnerability research, exploit validation, and security testing; Daybreak cybersecurity capabilities also went live on AWS via Amazon Bedrock in support of enterprise security workflows. (Link 1· Link 2)
- [Aug 11] OpenAI begins testing ads in ChatGPT: emphasizing clear labeling, answer independence, privacy protection, and user control. (Link)
- [Aug 11] Google publishes research on real-time clinical video consultation for its AMIE medical AI system: a first-of-its-kind study validating AMIE's real-time video consultation capability in simulated scenarios. (Link)
Note
This edition excludes customer case-study pieces (such as coverage of Zapier, Virgin Atlantic, RingCentral, and Model ML using ChatGPT Work) and personnel/commercial announcements (CRO appointment, ChatGPT Business seat promotions), treating them as vendor marketing content rather than "frontier developments."